None defined yet.
Continker is an Amsterdam studio building sovereign AI cloud platforms, applied AI, and local models that run offline. Infrastructure organizations own and run themselves, on European terms.
The infrastructure layer organizations own and run themselves. Model serving, retrieval, and agent workflows on hardware the customer controls, with audit and air-gap built in.
The systems that run on it. Language-model agents, retrieval, and automation built for real operational work.
Small, fine-tuned models that run offline on commodity hardware, so the model layer is owned too.
Sovereignty here is technical, not geographic. It is about who controls the keys, the runtime, and the operators, not where the data sits.
MetroLLM-Bench: a benchmark for running a transit kiosk from a prose prompt, with open-weight students that handle the task offline on commodity hardware. Published as arXiv:2609.10016 (September 2026). The 2.6 GB student matches GPT-5.4 at maximum reasoning effort on the deterministic tier and exceeds both GPT-5.6 tiers.
| Model | Size | Runs on |
|---|---|---|
| Qwen3.5-2B-metro-v24 | 1.2 GB | anything, including CPU-only |
| Qwen3.5-4B-metro-v24 | 2.6 GB | a 16 GB laptop |
| Qwen3.5-9B-metro-v24 | 5.3 GB | a 24 GB machine |
| Qwen3.5-27B-metro-v24 | 16 GB | a 32 GB machine |
Paper · Dataset (955 cases) · Live demo · Collection · Benchmark and code
More at continker.ai.
Transit kiosk run by a 9B local model (arXiv:2609.10016)
Transit kiosk run by a 9B local model (arXiv:2609.10016)
